Computer vision is the field of artificial intelligence that enables computers to interpret images and video: to recognize objects, faces and text, follow movement, and measure and reconstruct scenes in three dimensions. It powers applications from phone cameras and medical scans to self-driving cars, factory inspection and facial recognition.
From edges to deep learning
In 1966 MIT’s Seymour Papert proposed a “summer vision project” in which summer students would build programs to pick out and name objects in camera images – a task that ended up occupying the field for half a century. Early systems relied on hand-designed rules and features, such as edges and corners, and worked only in controlled settings. Statistical learning brought progress in the 2000s; the Viola–Jones detector of 2001 made real-time face detection possible and found its way into consumer cameras. The decisive change came with data and deep learning. Fei-Fei Li and colleagues built ImageNet, a dataset of millions of labeled images, and in 2012 the convolutional network AlexNet, trained on GPUs, sharply cut the error rate in the ImageNet competition. In 2015 Microsoft researchers reported a network that surpassed a widely cited estimate of human accuracy on that benchmark, and their ResNet won the competition later that year.
Vision today
Modern systems combine vision with language. OpenAI’s CLIP (2021) learned to connect images with text descriptions, and today’s multimodal models can describe photos, read handwriting, interpret charts and answer questions about what a camera sees. Image generators run the process in reverse (see Diffusion Model). In medicine, the US Food and Drug Administration has authorized more than 1,600 AI-enabled devices, about three-quarters of them in radiology; in agriculture, drones spot crop disease; in warehouses, robots pick items from shelves. Vision is also central to robotics and autonomous driving.
Surveillance and bias
The same technology enables mass surveillance. Facial recognition lets police identify people from photos, and Clearview AI scraped more than 30 billion photos from the web for a face-search service, drawing fines from European regulators. Joy Buolamwini and Timnit Gebru’s 2018 “Gender Shades” study found commercial gender classifiers far less accurate for darker-skinned women than for lighter-skinned men, A 2019 follow-up audit found that vendors had narrowed the gaps, and the study fed into debates that led some US cities, starting with San Francisco in 2019, to bar police use of facial recognition; US government testing finds much smaller demographic differences in the most accurate algorithms (see Algorithmic Bias). The EU AI Act prohibits untargeted scraping of facial images to build recognition databases and allows police to use real-time remote biometric identification in public spaces only under narrow exceptions.
Why it matters
Computer vision gives machines eyes, and with them the ability to help in the physical world – to read scans faster, make roads safer and describe the world to people who are blind or have low vision. Its power to identify and track individuals calls for clear limits on surveillance, independent accuracy testing and accountability for misuse, so the benefits can grow without eroding privacy.
Written with Claude, made by Anthropic, whose Claude models can analyze images.
Related
Deep Learning · Neural Network · Multimodal Model · Diffusion Model · GPU · Algorithmic Bias · Deepfake · Natural Language Processing
Sources: Seymour Papert: The Summer Vision Project (MIT AI Memo 100) (July 1966) · Paul Viola, Michael Jones: Rapid Object Detection using a Boosted Cascade of Simple Features (CVPR) (2001) · Jia Deng et al.: ImageNet: A large-scale hierarchical image database (CVPR) (2009) · Alex Krizhevsky, Ilya Sutskever, Geoffrey E. Hinton: ImageNet Classification with Deep Convolutional Neural Networks (NeurIPS) (2012) · Kaiming He et al.: Delving Deep into Rectifiers: Surpassing Human-Level Performance on ImageNet Classification (arXiv) (February 2015) · Kaiming He et al.: Deep Residual Learning for Image Recognition (arXiv) (December 2015) · Alec Radford et al.: Learning Transferable Visual Models From Natural Language Supervision (arXiv) (February 2021) · FDA: List of Artificial Intelligence-Enabled Medical Devices (updated June 2026) · Joy Buolamwini, Timnit Gebru: Gender Shades (Proceedings of Machine Learning Research) (2018) · Inioluwa Deborah Raji, Joy Buolamwini: Actionable Auditing (AIES) (2019) · Patrick Grother et al.: Face Recognition Vendor Test, Part 3: Demographic Effects (NISTIR 8280) (December 2019) · Dutch Data Protection Authority: Dutch DPA imposes a fine on Clearview because of illegal data collection for facial recognition (September 3, 2024) · EU AI Act: Article 5, Prohibited AI practices
Leave a Reply